Download The reform of SONIA - A supplementary

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Taylor's law wikipedia , lookup

Bootstrapping (statistics) wikipedia , lookup

Transcript
February 2017
The reform of SONIA
A supplementary consultation and an update on timing
February 2017
The reform of SONIA
A supplementary consultation and an update on timing
This Consultation Paper includes illustrative calculations, based on the Bank of England’s
Sterling Money Market data collection, which are intended only to show how the proposed
new SONIA methodology would work, and should not be used for any other purpose. The
proposed methodology may change before it is finalised, including in response to comments
made during the consultation process.
© Bank of England 2017
The reform of SONIA February 2017
3
1 Overview
1 The Bank of England set out its detailed proposals for
the reform of the SONIA interest rate benchmark, and how
they would be implemented, in a consultation issued in
October 2016 (hereafter the ‘2016 consultation’).(1)
2 This supplementary consultation revises the Bank’s
proposed approach on one specific issue relating to the
averaging methodology to be used for SONIA, and seeks
further feedback on the new proposal. As part of this it also
provides data on reformed SONIA up until 1 February 2017.
3 This document also provides an update on the
anticipated transition timetable.
1.1 Revised averaging methodology
4 The 2016 consultation proposed that SONIA be
calculated as the volume-weighted median rate of eligible
transactions. This proposal received generally supportive
feedback, but a combination of factors has prompted us to
reconsider our preferred averaging methodology.
5 In doing so we have reached the view that the trimmed
mean(2) – which was not considered as part of the earlier
consultation – strikes a better balance between the Bank’s
objectives and is preferred to the median. In particular:
 The trimmed mean is likely to behave a little more
like existing SONIA. Like the median, the trimmed
mean limits the impact of those transactions at
outlying rates, seen in the Sterling Money Market
(SMM) data, which may not be reflective of broader
market conditions. In both cases this helps to
maintain overall consistency with current SONIA.
However, daily moves in the trimmed mean are in
general likely to track moves in existing SONIA
slightly more closely. That may be most evident on
days when the underlying distribution of market
rates is more skewed than usual – for example as has
been the case at quarter-ends in recent years.
(1) See www.bankofengland.co.uk/markets/Documents/
soniareformcp1016.pdf
(2) The volume-weighted trimmed mean is calculated by removing
transactions at outlying rates and then calculating the volumeweighted mean of the remaining transactions.
 The trimmed mean is less sensitive to erroneous or
potentially manipulative transactions of plausible
size, or data reporting problems. In contrast,
relatively small changes in the distribution of
transactions can occasionally affect the median.
6 The following section explains in more detail the Bank’s
rationale for the switch in the proposed averaging
methodology. Box 1 compares the trimmed mean and the
median with current SONIA. It shows that, like the median,
the trimmed mean returns a rate a little more than 1 basis
point below current SONIA over the available sample
period.
7 Respondents are asked for feedback on the revised
proposal for averaging methodology by 16 March 2017 via
email to the Bank at:
[email protected]
1.2 Update on timetable for transition to reformed SONIA
8 The Bank now anticipates that the transition to
reformed SONIA will occur in March or April 2018, rather
than the end of 2017 as previously stated. The change in
timing reflects in part the additional time required for this
consultation, and in part responds to feedback received on
the proposed timeline set out in the 2016 consultation.
9 The Bank will provide a summary of feedback to this and
the 2016 consultation, together with its response, by endMarch 2017. In doing so the Bank will set out its final view
on the design of SONIA and the transition and publication
arrangements.
10 The revised implementation timetable therefore allows
around twelve months between confirmation of the final
design of reformed SONIA and its implementation.
4
The reform of SONIA February 2017
Daily changes in the proposed rate have tracked changes in current
SONIA closely over the sample period, with a high correlation
between the series (Charts C and D).
Box 1
The characteristics of reformed SONIA
This box outlines the illustrative statistical characteristics of
reformed SONIA, comparing the use of a volume-weighted
trimmed mean and a volume-weighted median.(a) For the purposes
of this analysis the eligibility criteria for input data proposed in the
2016 consultation have been used.
The data presented have been compiled using contributions to the
Bank’s Sterling Money Market (SMM) data collection, up to and
including 1 February 2017. For a provisional set of ‘early adopters’,
comprising roughly 80% of the full population, data are available
from 1 February 2016. The full reporting population has been
contributing since 1 July 2016.
Over the available sample both methodologies return a rate that is
on average a little more than 1 basis point below current SONIA
(Charts A and B), reflecting that some bilateral transactions in the
sample occur at slightly lower rates than brokered transactions.
Chart A Comparison of current and reformed SONIA using volumeweighted trimmed mean
Spread (left-hand scale)
Current SONIA (right-hand scale)
Percent
Reformed SONIA Trimmed Mean (right-hand scale)
0.50
Basis
points
6.0
4.0
0.30
2.0
SONIA is widely used as the reference rate in overnight index swap
contracts. In this context the behaviour of the compounded rate
may be of interest to users. Charts E and F show that, on a onemonth compounded basis, the volatility of reformed SONIA using
both methodologies is low and very similar to that of existing
SONIA.
The largest difference between the two methodologies is at yearend, when the trimmed mean fell by more than the median in
response to the negative skew in the transaction rates in the Bank’s
SMM data collection (Chart G). It should be noted that in similar
circumstances at the end of September the median fell by slightly
more than the trimmed mean.
Table 1 presents summary statistics comparing the two calculation
methodologies for reformed SONIA with current SONIA.
Chart B Comparison of current and reformed SONIA using volumeweighted median
Spread (left-hand scale)
Current SONIA (right-hand scale)
Reformed SONIA Median (right-hand scale)
Basis
points
6.0
4.0
0.30
2.0
0.10
0.0
0.10
0.0
-0.10
-2.0
-0.10
-2.0
-0.30
-4.0
-0.30
-4.0
Full population
-6.0
Feb-16
Apr-16
Jun-16
Aug-16
Oct-16
Full population
Dec-16
-0.50
Feb-17
Chart C Comparison of daily changes in current and reformed
SONIA using volume-weighted trimmed mean
Current SONIA
Reformed SONIA Trimmed Mean
Apr-16
Jun-16
Aug-16
Oct-16
-6.0
Feb-16
Dec-16
Apr-16
Jun-16
Aug-16
Oct-16
Dec-16
-0.50
Feb-17
Chart D Comparison of daily changes in current and reformed
SONIA using volume-weighted median
Current SONIA
Reformed SONIA Median
Pecent
0.15
Full population
Feb-16
Percent
0.50
Percent
0.15
0.10
0.10
0.05
0.05
0.00
0.00
-0.05
-0.05
-0.10
-0.10
-0.15
-0.15
-0.20
-0.20
-0.25
Feb-17
Full population
Feb-16
Apr-16
Jun-16
Aug-16
Oct-16
-0.25
Dec-16
The reform of SONIA February 2017
Chart E Comparison of daily changes of one-month compounded
rate for current and reformed SONIA using volume-weighted
trimmed mean (b)
Current SONIA
Reformed SONIA Trimmed Mean
5
Chart F Comparison of daily changes of one-month compounded
rate for current and reformed SONIA using volume-weighted
median (b)
Current SONIA
Reformed SONIA Median
Basis points
0.40
Basis points
0.40
0.20
0.20
0.00
0.00
-0.20
-0.20
-0.40
-0.40
-0.60
-0.60
-0.80
Full population
Mar-16
Jun-16
-0.80
Full population
-1.00
Sep-16
Dec-16
Mar-16
Chart G Reformed SONIA and percentiles
May-16
Jul-16
Sep-16
-1.00
Nov-16
Jan-17
Table 1 Indicative summary statistics of the characteristics of current
and reformed SONIA based on data since 1 July 2016
25th - 75th percentiles
10th - 90th percentiles
Median
Trimmed Mean
Trimmed
Current
Median
mean
SONIA
Percent
0.50
Average spread to SONIA (basis points)
-1.14
-1.25
-
Correlation with SONIA
1.00
1.00
-
Correlation of daily changes with daily changes
in SONIA
0.96
0.93
-
0.25
Standard deviation of daily changes (basis
points)
1.92
2.00
2.12
0.20
Standard deviation of daily changes, excluding
4 August Bank Rate change (basis points)
0.77
0.89
1.26
0.45
0.40
0.35
0.30
0.15
0.10
0.05
Full population
Feb-16
Apr-16
Jun-16
Aug-16
Oct-16
Dec-16
0.00
Feb-17
(a) Sources: Bank of England, Wholesale Market Brokers’ Association Ltd and Bank
calculations. All data presented in this box and throughout this consultation are
available on the Bank’s website at
www.bankofengland.co.uk/markets/Documents/soniareform0217.xlsx
(b) Note that there was an error in the equivalent chart in the 2016 consultation,
published as Chart G, which has now been corrected.
6
The reform of SONIA February 2017
2 Averaging methodology
1 In the October 2016 consultation, the Bank considered
three volume-weighted averaging methodologies for
reformed SONIA: the mean (as used for existing SONIA); the
median; and the iteratively trimmed mean.
2 These options were assessed against three main criteria:
transparency; consistency with the current measure of
SONIA; and sensitivity to erroneous or manipulative trades.
The mean and median were preferred to the iteratively
trimmed mean on the grounds of simplicity and
transparency. The median was preferred to the mean for
being robust to outliers and producing a rate more
consistent with existing SONIA, while being sufficiently
simple and sufficiently insensitive to erroneous or
potentially manipulative trades.
3 Respondents were generally supportive of the proposal
to adopt the volume-weighed median for reformed SONIA.
However one respondent noted that in circumstances
where the distribution of rates is dispersed – including for
example where there are gaps between the rates at which
significant volumes of transactions occur – relatively small
shifts in the distribution could result in relatively large
changes in the rate. The Bank recognised this issue in
raising the proposed republication threshold of reformed
SONIA from 1 to 2 basis points.
4 Further data and analysis have shown that there are
scenarios in which the volume-weighted median could
plausibly be affected by 2 or more basis points by reporting
errors. This would in practice be highly unlikely to occur,
since it would require both an exceptionally low volume of
transactions in the vicinity of the median and significant
reporting problems at one or more of the larger reporting
institutions.
5 More broadly, this analysis has reemphasised that the
properties of the volume-weighed median are somewhat
more sensitive than alternative methodologies to the
distribution of transacted rates. This distribution may
change over time, whether for idiosyncratic reasons such as
reporting period-ends or days when policy rates are
changed, or for unforeseen longer term reasons, reflecting
possible structural changes in the market.(3)
(3) For example, the dispersion of transacted rates for current SONIA,
measured by the interquartile range of the volume-weighted
distribution of rates on brokered transactions, has averaged around 5
basis points over the past 10 years. However in the final quarter of
2008 it averaged over 40 basis points.
6 We believe our judgement that the median is
‘sufficiently insensitive to erroneous or potentially
manipulative trades’ remains sound under the observed
distribution of transactions. However the considerations
noted above prompted us to reopen the question of
averaging methodology.
7 In doing so we have reconsidered the available options
for the averaging methodology, and included one additional
option. While the 2016 consultation considered the
iterative trimmed mean (but discounted it as too complex),
we now consider the simpler ‘static’ trimmed volumeweighted mean (hereafter ‘trimmed mean’).
8 The trimmed mean is calculated by sorting all eligible
transactions in rate order, lowest to highest; excluding
(‘trimming’) the highest and lowest 25% by volume of
transactions; and calculating the volume-weighted mean of
the remaining 50% (by volume) of transactions. Annex 1
provides an illustrative example.
9 Conceptually the trimmed mean can be thought of as
lying between the mean – in which all of the transactions
are included in the calculation – and the median – which
depends only on the rate of the transaction at the centre of
the distribution.
10 The Bank judges that, in the context of the SMM data,
trimming the outer quartiles (and calculating the volumeweighted mean rate of the remaining 50% of transactions
by volume) strikes the best balance between: simplicity;
limiting the impact of outliers; and reducing susceptibility to
small changes in the distribution of transactions.
11 The remainder of this section compares the
performance of the volume-weighted trimmed mean and
volume-weighted median methodologies. We use the same
criteria as for the 2016 consultation.
2.1 Transparency
12 Consistent with IOSCO Principle 9 (Transparency of
Benchmark Determination), the Bank considers stakeholder
understanding of the calculation methodology to be core to
the transparency of benchmark determination.
13 A volume-weighted trimmed mean is a commonly used
and understood statistical approach which can be
transparently explained.
The reform of SONIA February 2017
2.2 Consistency with current SONIA
14 When applied to the SMM data, both the volumeweighted trimmed mean and median are highly correlated
with current SONIA. Focusing on daily changes, the
correlation between the trimmed mean and current SONIA
is 0.96, compared to 0.93 for the median.
15 The 2016 consultation discussed the negative skew
exhibited in the Bank’s Sterling Money Market (SMM) data
and how this is reflected in the different averaging
methodologies. The median was preferred to the mean
because, by minimising the effect of SMM transactions at
outlying rates, it produces a rate more consistent with
existing SONIA.
7
to behave somewhat more like current SONIA than is the
median.(5)
2.3 Sensitivity to manipulative or erroneous trades
20 It is important to ensure that reformed SONIA is not
overly sensitive either to potential attempts to manipulate
the benchmark through off-market transactions, or to datareporting problems, for example where technical issues
lead to late reporting.
16 The trimmed-mean performs similarly to the median in
this regard, returning – like the median – average rates a
little over 1 basis point below current SONIA over the
available sample period.(4)
21 As noted in the 2016 consultation, the volume-weighted
median is generally sufficiently robust to being influenced
by individual transactions of plausible size. It is also
generally robust to the absence of particular reporters.
However the nature of the median, together with the
tendency for transacted rates to be whole numbers of basis
points, means that on a minority of days relatively small
changes in the distribution could have a significant impact
on the median.
17 However in circumstances where the distribution of
transacted rates is skewed for reasons that are reflective of
broader money market conditions, it may be desirable for
this to be captured in the averaging methodology, as it
would be for current SONIA. To take an example, it has
been common in recent years for sterling money market
rates at reporting-period ends to dip below their average
levels, reflecting among other factors a reduced supply of
bank balance sheet capacity. Table A compares the
December 2016 average and year-end values using the
different methodologies.
22 In general, the trimmed mean performs better for these
purposes than the median. For example, the frequency of
days on which an erroneous or potentially manipulative
transaction of a plausible size would affect the rate by a
basis point or more is much lower. A £1 billion trade added
to the distribution of rates observed each day in the
available sample, 3 basis point above or below the existing
median, would have impacted the median by 1 basis point
or more on a little under 10% of occasions. The trimmed
mean would not have been affected by as much as a basis
point on any of these days.
Table A Rates at year-end using different methodologies (%)
Methodology
Current SONIA
SMM: VW-median
SMM: VW-trimmed mean
Dec 2016 average
excluding year-end
0.2152
0.2000
0.2039
30 Dec 2016
0.1618
0.2000
0.1742
18 The fall in current SONIA reflects a number of brokered
transactions executed at rates significantly lower than
those typically seen in preceding days, and a consequent
negative skew. While the distribution of bilateral
transactions at year-end exhibits the same feature – a
greater than usual negative skew – the median based on
the SMM data remained at 20 basis points. The trimmed
mean fell by 3 basis points and in this instance behaved
more like current SONIA.
19 This illustrates the general point, reflected in the
correlations noted above, that the trimmed mean is likely
(4) If the underlying distribution of SMM rates were to become more
skewed in future, these measures could diverge.
23 Similar arguments apply in relation to the potential for
data reporting problems to affect the benchmark. Our
analysis suggests that both the volume-weighted median
and the volume-weighted trimmed mean were robust to
any three reporters failing to report on a typical day in our
sample period (i.e. once the missing data was included, the
proposed republication threshold of 2 basis points
difference was not breached).
24 However, in a very limited number of scenarios, it was
found that the impact of late submissions of reporters could
have a material impact on the calculation of the volumeweighted median; this is a natural consequence when there
is a low volume of transactions in the vicinity of the median.
We would expect these occasions to be extremely rare –
since they require both multiple reporting failures and a
relatively unusual distribution of transactions. We note
(5) Consistent with this, we note that the volume-weighted trimmed
mean of brokered transactions over the past 10 years has a smaller
average absolute spread to actual SONIA than does the volumeweighted median, and its daily changes are more highly correlated.
8
The reform of SONIA February 2017
that, even in these early stages of SMM data reporting,
there has been only one occasion so far where three
reporters failed to report on time; as the data collection
continues to bed in, we anticipate performance will
improve.
2.4 Conclusion
25 In conclusion, the Bank proposes to use a volumeweighted trimmed mean as the averaging methodology for
reformed SONIA. It is preferred to the volume-weighted
median and mean because it: is simple to understand;
produces a rate somewhat more consistent with the
existing measure of SONIA; is robust to outliers and
unrepresentative trades; and is less sensitive to erroneous
or potentially manipulative trades of plausible size, or the
late submission of data by a number of reporters.
28 Where respondents provide a response which includes
information to be treated as confidential, they should
clearly indicate the information provided which should be
treated as such. If a request for disclosure is subsequently
received, in accordance with access to information regimes
under the Freedom of Information Act 2000 or the Data
Protection Act 1998, respondents’ indications will be taken
into account, but no assurance can be given that
confidentiality can be maintained in all circumstances. An
automatic confidentiality disclaimer generated by a
respondent’s IT system on emails will not, of itself, be
treated as constituting notice that such respondent regards
any information supplied as confidential.
29 The Bank will provide its summary and response to
feedback received during this and the 2016 consultation in
March 2017. This response will include:
Question

Do you agree with the Bank’s view that a volumeweighted trimmed mean should be the averaging
methodology for reformed SONIA?


2.5 Next steps
26 Responses or any general queries relating to this
consultation are invited by 16 March 2017. These should be
sent to the Bank at the following email address:
[email protected]
27 Responses to this document are intended to be
published on the Bank’s website. Individual responses will
be unattributed, but a list of respondents will be published.
If respondents would like their comments to be treated as
confidential (and hence to not be published) they should
contact the Bank using the above email address, to discuss
this.



publication of individual responses to this and the
2016 consultation and a summary of these;
confirmation of the reformed SONIA calculation
methodology and definition;
confirmation of the data to be published alongside
the SONIA rate;
the contingency methodology to be used when
insufficient data is available;
an update of the indicative reformed SONIA data;
and
an update on the intended mechanics of transition.
30 Prior to the transition to reformed SONIA, the Bank will
publish additional information, in line with international
best practice for benchmark administration, including the
design of its SONIA governance arrangements.
The reform of SONIA February 2017
9
Annex Trimmed mean example
1 The volume-weighted trimmed mean that the Bank
proposes to use for reformed SONIA is calculated by sorting
all eligible transactions in rate order, lowest to highest;
excluding (‘trimming’) the highest and lowest 25% by
volume of transactions; and calculating the volume
weighted mean of the remaining 50% (by volume) of
transactions.
2 In the case that the volume of a single transaction spans
across the 25% (or 75%) threshold for trimming, that
transaction’s volume will be pro-rated, such that exactly
50% of the total eligible volume will be used in the volumeweighted mean calculation.
3 Figure A provides an example of how transaction
volume would be pro-rated in calculating the trimmed
mean. Each box in the column represents a single
transaction at a particular rate. The left column shows the
full population of transactions on a given day. On applying
the trim – in the right hand column – a portion of the
volume of transactions at rates of 0.23% and 0.25% falls
outside the middle 50% of the total volume and is
discarded, so that the calculation of the trimmed mean uses
only the middle 50% of the distribution.
Figure A An example of pro-rating the volume of transactions at
the edge of the middle 50%
Full population
Rate Volume
0.20%
0.21%
50
75
0.22%
100
Trimmed sample
Rate Volume
100%
75%
0.23%
200
0.23%
175
0.24%
250
0.24%
250
0.25%
75
0.25%
200
0.26%
0.27%
75
50
50%
________________________
________________________
£ 1000
£ 500
Volume-weighted trimmed mean: 0.2380%
Volume-weighted median: 0.2400%
25%
0%